01
1. Atomise the claim
Take the sentence you intend to publish and break it into the smallest units that could independently be wrong. A statement that a service has closed after demand halved contains at least four checkable units: that the service has closed, the date on which it did, that demand fell, and that the fall was one half by some stated measure over some stated period. The causal link between the last two and the first is a fifth unit and usually the weakest.
Atomising is what turns a vague worry into work you can finish. It also shows you which parts you cannot check at all, which is valuable information at the drafting stage rather than an hour before deadline. Number the units and keep them; they become the rows of your verification log and the basis of any later correction.
02
2. Rank the evidence
Evidence is not equal. A primary record made by the body that took the action outranks a description of that record, which outranks somebody's recollection of the description. For Irish reporting, the useful primary layer usually includes statutory instruments and Acts as published, official statistical releases from the Central Statistics Office, filings held in public registers, minutes and agendas of public bodies, court lists and published decisions.
A press release is a secondary source about a primary one, and it is written to be quoted. Use it to locate the document it summarises, then read that. Where the primary record does not exist or cannot be obtained, say in the text which layer of evidence you actually have. Readers can weigh a sentence that names its evidence; they cannot weigh a sentence that hides it.
03
3. Reading official records properly
Records have grammar. A statistical release has a reference period, a revision policy, a definition section and often a note explaining a break in the series. A register entry has a filing date that is not the event date. A court list carries charges as alleged, not findings. A published decision has an operative part and a reasoning part, and quoting the reasoning as the outcome is a common and serious error.
Read the notes before the numbers. Most misreporting of official data comes from skipping the definitions: counting cases when the release counts episodes, comparing a provisional figure with a final one, or treating a rate as a count. Where a release offers a seasonally adjusted and an unadjusted series, say which you used. Requests under the Freedom of Information Act 2014 can obtain records that are not published, on the terms and with the exemptions that Act sets out.
04
4. Numbers that mislead honestly
Four arithmetic traps account for most number errors in copy. The first is confusing percentage change with percentage-point change. The second is comparing counts between populations of different size instead of using rates. The third is the moving base, where a fall of twenty per cent followed by a rise of twenty per cent does not return to the starting value. The fourth is the small denominator, where a change of two in a total of five becomes a dramatic percentage.
The remedy is mechanical. Write the arithmetic out in the margin of your notes with the units attached, state the base and the period in the sentence itself, and give the absolute figure alongside any percentage. If a number cannot be expressed with its base and period in one clause, it usually needs its own sentence rather than a bracket.
05
5. Images, video and the checks that scale
Visual material carries the strongest impression and the weakest provenance. The basic sequence is to establish where the file came from, whether an earlier version of it exists, whether the depicted place matches independent imagery of that place, and whether the depicted conditions match the record. Shadows, signage, vehicle plates, shopfronts and language on posters all carry checkable information.
Weather and daylight are the two most useful independent checks available to any reporter, because both are recorded and neither is easy to fake consistently. If the file shows heavy rain on an afternoon that was recorded as dry, either the date or the file is wrong. Where a file cannot be placed or dated, it is not publishable as evidence of anything, however striking it looks.
06
6. The two-source rule and its limits
Two independent sources are better than one, but independence is the word that does the work. Two people repeating the same briefing are one source. Two publications citing each other are one source with an extra step. Before counting a second source, ask how that person or record came to know the thing, and whether that route is genuinely separate from the first.
The rule also cannot rescue a claim that no available source could settle. A matter of internal intention, a private conversation with no record, or a statistic nobody collects will not become checkable through repetition. In those cases the honest choice is to report the dispute rather than the fact: who says what, on what basis, and what remains unknown.
07
7. Keeping a verification log
A verification log is one row per atomised claim, recording the claim, the evidence sought, what was found, the identifier of the document or the timestamp of the recording, the date checked and the outcome. It takes minutes to keep and saves hours when a complaint arrives, because it converts a memory contest into a documented sequence.
The log is also a drafting tool. Any row marked unresolved should correspond to a hedge in the text, and any hedge in the text should correspond to a row in the log. If a sentence in your draft is confident and the log says the claim is unverified, one of the two is wrong. Keep the log with the notes and the audio, under the same retention and confidentiality rules.
08
8. Flagging what cannot be confirmed
Unverified material is publishable in some circumstances and not in others, and the deciding factors are the seriousness of the claim, the public interest in reporting it, and whether the uncertainty can be described accurately. What is never acceptable is a confident sentence resting on an unchecked claim, with the doubt moved to a paragraph a reader may never reach.
Language should match evidence precisely. There is a real difference between a document showing something, a person stating it, a person alleging it, and a reporter being unable to establish it. Where a right of reply has been sought and not received, say when it was sought and by what means. Where a claim was checked and failed, saying so is itself reporting, and often the most useful sentence in the piece.
Claim types and the evidence they need
| Claim type | Evidence that settles it | Common failure |
|---|---|---|
| An event happened | Contemporaneous record, official confirmation, first-hand account | Relying on a later summary of the event |
| A quantity changed | Published series with definitions and reference period | Mixing provisional and final figures |
| A person holds a role | Register, official listing, organisation's own current record | Using an outdated staff page |
| A document says something | The document itself, cited by section | Quoting reasoning as if it were the decision |
| One thing caused another | Rarely settled; needs stated basis and named authority | Publishing the causal link as established fact |
Checklist before a claim goes into copy
- Claim split into units, each with its own evidence line.
- Primary record obtained, or the evidence level stated in the text.
- Definitions, reference period and revision status read before the numbers.
- Arithmetic written out with units, base and period.
- Second source genuinely independent of the first.
- Verification log row for every hedge, and a hedge for every unresolved row.
Questions readers ask
Is a government press release a primary source?
No. It is a secondary account of a decision, dataset or document, written to be quoted. Use it to find the underlying record, then cite that.
How many sources are enough?
There is no fixed number. Two genuinely independent sources are a working minimum for a serious contested claim, and no number of dependent sources adds anything.
What if a claim cannot be checked at all?
Report the dispute rather than the fact: who says what, on what basis, and what remains unknown. Do not move the doubt to the end of the story.